IISc Releases SraVaani AI Model for 65 Indian Languages

Researchers at IISc’s SPIRE Lab have released SraVaani, an open-source speech recognition model supporting 20 scheduled and 45 regional Indian languages and dialects.

The model was trained on the Vaani dataset, which includes over 31,000 hours of speech recorded from 156,000 people across 165 districts in 28 states.

SraVaani is available on Hugging Face under an MIT licence and achieved a 9.5 per cent word error rate on the Garo language during benchmark testing.